1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
18 #include "por-file-writer.h"
29 #include <data/case.h>
30 #include <data/casewriter-provider.h>
31 #include <data/casewriter.h>
32 #include <data/dictionary.h>
33 #include <data/file-handle-def.h>
34 #include <data/file-name.h>
35 #include <data/format.h>
36 #include <data/missing-values.h>
37 #include <data/short-names.h>
38 #include <data/value-labels.h>
39 #include <data/variable.h>
41 #include <libpspp/alloc.h>
42 #include <libpspp/hash.h>
43 #include <libpspp/message.h>
44 #include <libpspp/misc.h>
45 #include <libpspp/str.h>
46 #include <libpspp/version.h>
49 #define _(msgid) gettext (msgid)
51 /* Maximum width of a variable in a portable file. */
52 #define MAX_POR_WIDTH 255
54 /* Portable file writer. */
57 struct file_handle *fh; /* File handle. */
58 FILE *file; /* File stream. */
60 int lc; /* Number of characters on this line so far. */
62 size_t var_cnt; /* Number of variables. */
63 struct pfm_var *vars; /* Variables. */
65 int digits; /* Digits of precision. */
68 /* A variable to write to the portable file. */
71 int width; /* 0=numeric, otherwise string var width. */
72 int fv; /* Starting case index. */
75 static struct casewriter_class por_file_casewriter_class;
77 static bool close_writer (struct pfm_writer *);
78 static void buf_write (struct pfm_writer *, const void *, size_t);
79 static void write_header (struct pfm_writer *);
80 static void write_version_data (struct pfm_writer *);
81 static void write_variables (struct pfm_writer *, struct dictionary *);
82 static void write_value_labels (struct pfm_writer *,
83 const struct dictionary *);
84 static void write_documents (struct pfm_writer *,
85 const struct dictionary *);
87 static void format_trig_double (long double, int base_10_precision, char[]);
88 static char *format_trig_int (int, bool force_sign, char[]);
90 /* Returns default options for writing a portable file. */
91 struct pfm_write_options
92 pfm_writer_default_options (void)
94 struct pfm_write_options opts;
95 opts.create_writeable = true;
97 opts.digits = DBL_DIG;
101 /* Writes the dictionary DICT to portable file HANDLE according
102 to the given OPTS. Returns nonzero only if successful. DICT
103 will not be modified, except to assign short names. */
105 pfm_open_writer (struct file_handle *fh, struct dictionary *dict,
106 struct pfm_write_options opts)
108 struct pfm_writer *w = NULL;
113 /* Open file handle. */
114 if (!fh_open (fh, FH_REF_FILE, "portable file", "we"))
118 mode = S_IRUSR | S_IRGRP | S_IROTH;
119 if (opts.create_writeable)
120 mode |= S_IWUSR | S_IWGRP | S_IWOTH;
121 file = create_stream (fh_get_file_name (fh), "w", mode);
124 fh_close (fh, "portable file", "we");
125 msg (ME, _("An error occurred while opening \"%s\" for writing "
126 "as a portable file: %s."),
127 fh_get_file_name (fh), strerror (errno));
131 /* Initialize data structures. */
132 w = xmalloc (sizeof *w);
140 w->var_cnt = dict_get_var_cnt (dict);
141 w->vars = xnmalloc (w->var_cnt, sizeof *w->vars);
142 for (i = 0; i < w->var_cnt; i++)
144 const struct variable *dv = dict_get_var (dict, i);
145 struct pfm_var *pv = &w->vars[i];
146 pv->width = MIN (var_get_width (dv), MAX_POR_WIDTH);
147 pv->fv = var_get_case_index (dv);
150 w->digits = opts.digits;
153 msg (ME, _("Invalid decimal digits count %d. Treating as %d."),
158 /* Write file header. */
160 write_version_data (w);
161 write_variables (w, dict);
162 write_value_labels (w, dict);
163 if (dict_get_document_line_cnt (dict) > 0)
164 write_documents (w, dict);
165 buf_write (w, "F", 1);
166 if (ferror (w->file))
171 return casewriter_create (dict_get_next_value_idx (dict),
172 &por_file_casewriter_class, w);
175 /* Write NBYTES starting at BUF to the portable file represented by
176 H. Break lines properly every 80 characters. */
178 buf_write (struct pfm_writer *w, const void *buf_, size_t nbytes)
180 const char *buf = buf_;
182 if (ferror (w->file))
185 assert (buf != NULL);
186 while (nbytes + w->lc >= 80)
188 size_t n = 80 - w->lc;
191 fwrite (buf, n, 1, w->file);
192 fwrite ("\r\n", 2, 1, w->file);
198 fwrite (buf, nbytes, 1, w->file);
203 /* Write D to the portable file as a floating-point field. */
205 write_float (struct pfm_writer *w, double d)
208 format_trig_double (d, floor (d) == d ? DBL_DIG : w->digits, buffer);
209 buf_write (w, buffer, strlen (buffer));
210 buf_write (w, "/", 1);
213 /* Write N to the portable file as an integer field. */
215 write_int (struct pfm_writer *w, int n)
218 format_trig_int (n, false, buffer);
219 buf_write (w, buffer, strlen (buffer));
220 buf_write (w, "/", 1);
223 /* Write S to the portable file as a string field. */
225 write_string (struct pfm_writer *w, const char *s)
227 size_t n = strlen (s);
228 write_int (w, (int) n);
232 /* Write file header. */
234 write_header (struct pfm_writer *w)
236 static const char spss2ascii[256] =
238 "0000000000000000000000000000000000000000000000000000000000000000"
239 "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz ."
240 "<(+|&[]!$*);^-/|,%_>?`:$@'=\"000000~-0000123456789000-()0{}\\00000"
241 "0000000000000000000000000000000000000000000000000000000000000000"
245 for (i = 0; i < 5; i++)
246 buf_write (w, "ASCII SPSS PORT FILE ", 40);
248 buf_write (w, spss2ascii, 256);
249 buf_write (w, "SPSSPORT", 8);
252 /* Writes version, date, and identification records. */
254 write_version_data (struct pfm_writer *w)
262 if ((time_t) -1 == time (&t))
264 tm.tm_sec = tm.tm_min = tm.tm_hour = tm.tm_mon = tm.tm_year = 0;
269 tmp = localtime (&t);
271 sprintf (date_str, "%04d%02d%02d",
272 tmp->tm_year + 1900, tmp->tm_mon + 1, tmp->tm_mday);
273 sprintf (time_str, "%02d%02d%02d", tmp->tm_hour, tmp->tm_min, tmp->tm_sec);
274 buf_write (w, "A", 1);
275 write_string (w, date_str);
276 write_string (w, time_str);
278 /* Product identification. */
279 buf_write (w, "1", 1);
280 write_string (w, version);
282 /* Subproduct identification. */
283 buf_write (w, "3", 1);
284 write_string (w, host_system);
287 /* Write format F to file H. The format is first resized to fit
288 a value of the given WIDTH, which is handy in case F
289 represents a string longer than 255 bytes and thus WIDTH is
290 truncated to 255 bytes. */
292 write_format (struct pfm_writer *w, struct fmt_spec f, int width)
294 fmt_resize (&f, width);
295 write_int (w, fmt_to_io (f.type));
300 /* Write value V for variable VV to file H. */
302 write_value (struct pfm_writer *w, union value *v, struct variable *vv)
304 if (var_is_numeric (vv))
305 write_float (w, v->f);
308 int width = MIN (var_get_width (vv), MAX_POR_WIDTH);
309 write_int (w, width);
310 buf_write (w, v->s, width);
314 /* Write variable records. */
316 write_variables (struct pfm_writer *w, struct dictionary *dict)
320 short_names_assign (dict);
322 if (dict_get_weight (dict) != NULL)
324 buf_write (w, "6", 1);
325 write_string (w, var_get_short_name (dict_get_weight (dict), 0));
328 buf_write (w, "4", 1);
329 write_int (w, dict_get_var_cnt (dict));
332 for (i = 0; i < dict_get_var_cnt (dict); i++)
334 struct variable *v = dict_get_var (dict, i);
335 struct missing_values mv;
336 int width = MIN (var_get_width (v), MAX_POR_WIDTH);
338 buf_write (w, "7", 1);
339 write_int (w, width);
340 write_string (w, var_get_short_name (v, 0));
341 write_format (w, *var_get_print_format (v), width);
342 write_format (w, *var_get_write_format (v), width);
344 /* Write missing values. */
345 mv_copy (&mv, var_get_missing_values (v));
346 while (mv_has_range (&mv))
349 mv_pop_range (&mv, &x, &y);
352 buf_write (w, "9", 1);
355 else if (y == HIGHEST)
357 buf_write (w, "A", 1);
362 buf_write (w, "B", 1);
367 while (mv_has_value (&mv))
370 mv_pop_value (&mv, &value);
371 buf_write (w, "8", 1);
372 write_value (w, &value, v);
375 /* Write variable label. */
376 if (var_get_label (v) != NULL)
378 buf_write (w, "C", 1);
379 write_string (w, var_get_label (v));
384 /* Write value labels to disk. FIXME: Inefficient. */
386 write_value_labels (struct pfm_writer *w, const struct dictionary *dict)
390 for (i = 0; i < dict_get_var_cnt (dict); i++)
392 struct val_labs_iterator *j;
393 struct variable *v = dict_get_var (dict, i);
394 const struct val_labs *val_labs = var_get_value_labels (v);
397 if (val_labs == NULL)
400 buf_write (w, "D", 1);
402 write_string (w, var_get_short_name (v, 0));
403 write_int (w, val_labs_count (val_labs));
405 for (vl = val_labs_first_sorted (val_labs, &j); vl != NULL;
406 vl = val_labs_next (val_labs, &j))
408 write_value (w, &vl->value, v);
409 write_string (w, vl->label);
414 /* Write documents in DICT to portable file W. */
416 write_documents (struct pfm_writer *w, const struct dictionary *dict)
418 size_t line_cnt = dict_get_document_line_cnt (dict);
419 struct string line = DS_EMPTY_INITIALIZER;
422 buf_write (w, "E", 1);
423 write_int (w, line_cnt);
424 for (i = 0; i < line_cnt; i++)
426 dict_get_document_line (dict, i, &line);
427 write_string (w, ds_cstr (&line));
432 /* Writes case C to the portable file represented by WRITER. */
434 por_file_casewriter_write (struct casewriter *writer, void *w_,
437 struct pfm_writer *w = w_;
440 if (!ferror (w->file))
442 for (i = 0; i < w->var_cnt; i++)
444 struct pfm_var *v = &w->vars[i];
447 write_float (w, case_num_idx (c, v->fv));
450 write_int (w, v->width);
451 buf_write (w, case_str_idx (c, v->fv), v->width);
456 casewriter_force_error (writer);
462 por_file_casewriter_destroy (struct casewriter *writer, void *w_)
464 struct pfm_writer *w = w_;
465 if (!close_writer (w))
466 casewriter_force_error (writer);
469 /* Closes a portable file after we're done with it.
470 Returns true if successful, false if an I/O error occurred. */
472 close_writer (struct pfm_writer *w)
483 memset (buf, 'Z', sizeof buf);
484 buf_write (w, buf, w->lc >= 80 ? 80 : 80 - w->lc);
486 ok = !ferror (w->file);
487 if (fclose (w->file) == EOF)
491 msg (ME, _("An I/O error occurred writing portable file \"%s\"."),
492 fh_get_file_name (w->fh));
495 fh_close (w->fh, "portable file", "we");
503 /* Base-30 conversion.
505 Portable files represent numbers in base-30 format, so we need
506 to be able to convert real and integer number to that base.
507 Older versions of PSPP used libgmp to do so, but this added a
508 big library dependency to do just one thing. Now we do it
509 ourselves internally.
511 Important fact: base 30 is called "trigesimal". */
513 /* Conversion base. */
514 #define BASE 30 /* As an integer. */
515 #define LDBASE ((long double) BASE) /* As a long double. */
517 /* This is floor(log30(2**31)), the minimum number of trigesimal
518 digits that a `long int' can hold. */
521 /* pow_tab[i] = pow (30, pow (2, i)) */
522 static long double pow_tab[16];
524 /* Initializes pow_tab[]. */
528 static bool did_init = false;
532 /* Only initialize once. */
537 /* Set each element of pow_tab[] until we run out of numerical
540 for (power = 30.0L; power < DBL_MAX; power *= power)
542 assert (i < sizeof pow_tab / sizeof *pow_tab);
543 pow_tab[i++] = power;
547 /* Returns 30**EXPONENT, for 0 <= EXPONENT <= log30(DBL_MAX). */
549 pow30_nonnegative (int exponent)
554 assert (exponent >= 0);
555 assert (exponent < 1L << (sizeof pow_tab / sizeof *pow_tab));
558 for (i = 0; exponent > 0; exponent >>= 1, i++)
565 /* Returns 30**EXPONENT, for log30(DBL_MIN) <= EXPONENT <=
571 return pow30_nonnegative (exponent);
573 return 1.L / pow30_nonnegative (-exponent);
576 /* Returns the character corresponding to TRIG. */
578 trig_to_char (int trig)
580 assert (trig >= 0 && trig < 30);
581 return "0123456789ABCDEFGHIJKLMNOPQRST"[trig];
584 /* Formats the TRIG_CNT trigs in TRIGS[], writing them as
585 null-terminated STRING. The trigesimal point is inserted
586 after TRIG_PLACES characters have been printed, if necessary
587 adding extra zeros at either end for correctness. Returns the
588 character after the formatted number. */
590 format_trig_digits (char *string,
591 const char trigs[], int trig_cnt, int trig_places)
596 while (trig_places++ < 0)
600 while (trig_cnt-- > 0)
602 if (trig_places-- == 0)
604 *string++ = trig_to_char (*trigs++);
606 while (trig_places-- > 0)
612 /* Helper function for format_trig_int() that formats VALUE as a
613 trigesimal integer at CP. VALUE must be nonnegative.
614 Returns the character following the formatted integer. */
616 recurse_format_trig_int (char *cp, int value)
618 int trig = value % BASE;
621 cp = recurse_format_trig_int (cp, value);
622 *cp++ = trig_to_char (trig);
626 /* Formats VALUE as a trigesimal integer in null-terminated
627 STRING[]. VALUE must be in the range -DBL_MAX...DBL_MAX. If
628 FORCE_SIGN is true, a sign is always inserted; otherwise, a
629 sign is only inserted if VALUE is negative. */
631 format_trig_int (int value, bool force_sign, char string[])
642 /* Format integer. */
643 string = recurse_format_trig_int (string, value);
648 /* Determines whether the TRIG_CNT trigesimals in TRIGS[] warrant
649 rounding up or down. Returns true if TRIGS[] represents a
650 value greater than half, false if less than half. If TRIGS[]
651 is exactly half, examines TRIGS[-1] and returns true if odd,
652 false if even ("round to even"). */
654 should_round_up (const char trigs[], int trig_cnt)
656 assert (trig_cnt > 0);
658 if (*trigs < BASE / 2)
660 /* Less than half: round down. */
663 else if (*trigs > BASE / 2)
665 /* Greater than half: round up. */
670 /* Approximately half: look more closely. */
672 for (i = 1; i < trig_cnt; i++)
675 /* Slightly greater than half: round up. */
679 /* Exactly half: round to even. */
680 return trigs[-1] % 2;
684 /* Rounds up the rightmost trig in the TRIG_CNT trigs in TRIGS[],
685 carrying to the left as necessary. Returns true if
686 successful, false on failure (due to a carry out of the
687 leftmost position). */
689 try_round_up (char *trigs, int trig_cnt)
693 char *round_trig = trigs + --trig_cnt;
694 if (*round_trig != BASE - 1)
696 /* Round this trig up to the next value. */
701 /* Carry over to the next trig to the left. */
705 /* Ran out of trigs to carry. */
709 /* Converts VALUE to trigesimal format in string OUTPUT[] with the
710 equivalent of at least BASE_10_PRECISION decimal digits of
711 precision. The output format may use conventional or
712 scientific notation. Missing, infinite, and extreme values
713 are represented with "*.". */
715 format_trig_double (long double value, int base_10_precision, char output[])
717 /* Original VALUE was negative? */
720 /* Number of significant trigesimals. */
721 int base_30_precision;
723 /* Base-2 significand and exponent for original VALUE. */
727 /* VALUE as a set of trigesimals. */
728 char buffer[DBL_DIG + 16];
732 /* Number of trigesimal places for trigs.
733 trigs[0] has coefficient 30**(trig_places - 1),
734 trigs[1] has coefficient 30**(trig_places - 2),
736 In other words, the trigesimal point is just before trigs[0].
740 /* Number of trigesimal places left to write into BUFFER. */
745 /* Handle special cases. */
751 /* Make VALUE positive. */
760 /* Adjust VALUE to roughly 30**3, by shifting the trigesimal
761 point left or right as necessary. We approximate the
762 base-30 exponent by obtaining the base-2 exponent, then
763 multiplying by log30(2). This approximation is sufficient
764 to ensure that the adjusted VALUE is always in the range
765 0...30**6, an invariant of the loop below. */
767 base_2_sig = frexp (value, &base_2_exp);
768 if (errno != 0 || !finite (base_2_sig))
770 if (base_2_exp == 0 && base_2_sig == 0.)
772 if (base_2_exp <= INT_MIN / 20379L || base_2_exp >= INT_MAX / 20379L)
774 trig_places = (base_2_exp * 20379L / 100000L) + CHUNK_SIZE / 2;
775 value *= pow30 (CHUNK_SIZE - trig_places);
777 /* Dump all the trigs to buffer[], CHUNK_SIZE at a time. */
780 for (trigs_to_output = DIV_RND_UP (DBL_DIG * 2, 3) + 1 + (CHUNK_SIZE / 2);
782 trigs_to_output -= CHUNK_SIZE)
787 /* The current chunk is just the integer part of VALUE,
788 truncated to the nearest integer. The chunk fits in a
791 assert (pow30 (CHUNK_SIZE) <= LONG_MAX);
792 assert (chunk >= 0 && chunk < pow30 (CHUNK_SIZE));
796 /* Append the chunk, in base 30, to trigs[]. */
797 for (trigs_left = CHUNK_SIZE; chunk > 0 && trigs_left > 0; )
799 trigs[trig_cnt + --trigs_left] = chunk % 30;
802 while (trigs_left > 0)
803 trigs[trig_cnt + --trigs_left] = 0;
804 trig_cnt += CHUNK_SIZE;
806 /* Proceed to the next chunk. */
809 value *= pow (LDBASE, CHUNK_SIZE);
812 /* Strip leading zeros. */
813 while (trig_cnt > 1 && *trigs == 0)
820 /* Round to requested precision, conservatively estimating the
821 required base-30 precision as 2/3 of the base-10 precision
822 (log30(10) = .68). */
823 assert (base_10_precision > 0);
824 if (base_10_precision > LDBL_DIG)
825 base_10_precision = LDBL_DIG;
826 base_30_precision = DIV_RND_UP (base_10_precision * 2, 3);
827 if (trig_cnt > base_30_precision)
829 if (should_round_up (trigs + base_30_precision,
830 trig_cnt - base_30_precision))
832 /* Try to round up. */
833 if (try_round_up (trigs, base_30_precision))
835 /* Rounding up worked. */
836 trig_cnt = base_30_precision;
840 /* Couldn't round up because we ran out of trigs to
841 carry into. Do the carry here instead. */
850 trig_cnt = base_30_precision;
855 /* No rounding required: fewer digits available than
859 /* Strip trailing zeros. */
860 while (trig_cnt > 1 && trigs[trig_cnt - 1] == 0)
866 if (trig_places >= -1 && trig_places < trig_cnt + 3)
868 /* Use conventional notation. */
869 format_trig_digits (output, trigs, trig_cnt, trig_places);
873 /* Use scientific notation. */
875 op = format_trig_digits (output, trigs, trig_cnt, trig_cnt);
876 op = format_trig_int (trig_places - trig_cnt, true, op);
881 strcpy (output, "0");
885 strcpy (output, "*.");
889 static struct casewriter_class por_file_casewriter_class =
891 por_file_casewriter_write,
892 por_file_casewriter_destroy,